Agent Scrum: Running a team of AI coding agents in two-hour sprints (daveremy.github.io)

Agent Scrum introduces a novel sprint methodology for organizing AI coding agents in dynamic software development environments. Over 38 days, a fleet of AI agents conducted 324 two-hour sprints, revealing that 79% of committed tasks were completed effectively. The method adapts traditional Scrum principles—like time-boxed sprints and defined roles—to suit AI's capabilities, allowing for rapid, continuous delivery without the typical overhead of human coordination. While sprints are short and duration is measured by token velocity, the system still faced challenges in compounding knowledge, with many retrospective actions remaining open. This development is significant for the AI/ML community as it not only enhances the efficiency of coding agents but also provides a framework to systematically scale AI collaboration. By maintaining a human oversight role and using a structured approach for planning, the method ensures accountability while optimizing execution and context management across multiple agents. The results suggest that AI-coded software factories can rival human teams in productivity without succumbing to common pitfalls of traditional Scrum, indicating a promising direction for integrating AI into agile software development practices.